def test_image_contrast(): """Test: increases the contrast in order to make possible fourier transforms smoother""" #inputs for functions image = Image.open('../data/protein_matrix_image.png') contrast_value = 2 #function performance enhanced_image = preprocessing.image_contrast(image, contrast_value) #asserts and checks return enhanced_image
def test_objectnumber(): """Test: Gets the object number for the global and local thresholds""" #inputs for functions image = Image.open('../data/protein_matrix_image.png') desired_color = 'b' #processing split = preprocessing.color_split_image(image, desired_color) contrast_image = preprocessing.image_contrast(split, 2) block_size = 15 #running the functions objects.objectnumber(contrast_image, block_size) #asserts and checks return
def test_distance_histograms(): """Test: obtaining 10-bin histograms of centroid distances""" #inputs for functions image = Image.open('../data/protein_matrix_image.png') desired_color = 'b' #processing split = preprocessing.color_split_image(image, desired_color) contrast_image = preprocessing.image_contrast(split, 2) block_size = 15 #running the functions objects.distance_histograms(contrast_image, block_size) #asserts and checks return
def test_objectcentroids(): """Test: obtaining the object centroids""" #inputs for functions image = Image.open('../data/protein_matrix_image.png') desired_color = 'b' #processing split = preprocessing.color_split_image(image, desired_color) contrast_image = preprocessing.image_contrast(split, 2) block_size = 15 #running the functions object_centroids, object_centroids_local = objects.objectcentroids( contrast_image, block_size) #asserts and checks return
def test_stats(): """Test: gets basic stats for whatever property input""" #inputs for functions image = Image.open('../data/protein_matrix_image.png') desired_color = 'b' #processing split = preprocessing.color_split_image(image, desired_color) image = preprocessing.image_contrast(split, 2) #inputs for functions #labels otsu_global_labels = objects.global_labels(image) properties_global = measure.regionprops(otsu_global_labels) areas_global = [prop.bbox_area for prop in properties_global] #final inputs property = areas_global #running the functions objects.stats(property) #asserts and checks return
def test_distancesarr(): """Test: gets the distances between image and objects""" #inputs for functions image = Image.open('../data/protein_matrix_image.png') desired_color = 'b' #processing split = preprocessing.color_split_image(image, desired_color) image = preprocessing.image_contrast(split, 2) block_size = 15 #final inputs otsu_global_labels = objects.global_labels(image) properties_global = measure.regionprops(otsu_global_labels) image_centroid = properties_global[0].centroid global_binarys = objects.global_binary(image) object_centroids = feature.blob_log(global_binarys) row = [3, 3] #running the functions distances = objects.distancesarr(image_centroid, object_centroids) #asserts and checks return